1  # Threshold-Triggered Automation
  2  
  3  *proto-026 | Operational pattern for spawning agents when metrics exceed thresholds*
  4  
  5  ---
  6  
  7  - **principle**
  8    - "When a metric exceeds a threshold, spawn an agent to address it automatically."
  9    - "Don't wait for human intervention on predictable triggers."
 10  
 11  - **shape**
 12    - Define thresholds for each issue type
 13    - Monitor continuously (or on-demand)
 14    - When threshold crossed → spawn appropriate agent
 15    - Agent runs with same safeguards as manual invocation
 16    - Results flow back for review
 17  
 18  ---
 19  
 20  ## The Insight
 21  
 22  Manual intervention doesn't scale. When you know:
 23  1. What metric to watch
 24  2. What threshold indicates action needed
 25  3. What agent can address it
 26  
 27  ...then waiting for a human to notice and act is waste.
 28  
 29  ```
 30  MANUAL PATTERN:
 31  Human notices → Human decides → Human spawns agent
 32      ↑
 33      └── bottleneck, delay, missed issues
 34  
 35  THRESHOLD-TRIGGERED PATTERN:
 36  System monitors → Threshold crossed → Agent spawns automatically
 37                                             ↓
 38                                      Human reviews results
 39  ```
 40  
 41  The human still has oversight—they review results and can adjust thresholds—but the system doesn't wait.
 42  
 43  ---
 44  
 45  ## Implementation
 46  
 47  ```python
 48  # Threshold configuration
 49  THRESHOLDS = {
 50      "phoenix_update": 10,      # 10+ violations = spawn fixer
 51      "hygiene_check": 5,        # 5+ violations = spawn fixer
 52      "alignment_report": 20,    # 20+ missing = spawn fixer
 53  }
 54  
 55  # Strategy per issue type
 56  STRATEGIES = {
 57      "phoenix_update": "infrastructure",  # Build enforceable code
 58      "hygiene_check": "infrastructure",
 59      "alignment_report": "protocol",      # Update documentation
 60  }
 61  
 62  def check_thresholds(issues_by_type: Dict) -> List[str]:
 63      """Return issue types that exceed thresholds."""
 64      triggers = []
 65      for issue_type, threshold in THRESHOLDS.items():
 66          if len(issues_by_type.get(issue_type, [])) >= threshold:
 67              triggers.append(issue_type)
 68      return triggers
 69  
 70  def spawn_for_triggers(triggers: List[str], issues: Dict) -> List[Agent]:
 71      """Spawn agents for triggered thresholds."""
 72      agents = []
 73      for issue_type in triggers:
 74          agent = create_agent(
 75              issue_type=issue_type,
 76              strategy=STRATEGIES[issue_type],
 77              violations=issues[issue_type],
 78          )
 79          agents.append(agent)
 80      return agents
 81  ```
 82  
 83  ---
 84  
 85  ## When to Apply
 86  
 87  | Signal | Action |
 88  |--------|--------|
 89  | Violation count > threshold | Spawn fixer agent |
 90  | Error rate > threshold | Spawn diagnostic agent |
 91  | Staleness > threshold | Spawn refresh agent |
 92  | Drift > threshold | Spawn alignment agent |
 93  
 94  **The key insight:** If you can define the threshold and the response, automate it.
 95  
 96  ---
 97  
 98  ## Safeguards
 99  
100  Automated spawning requires same safeguards as manual:
101  
102  1. **Confidence decay** - Agent stops if confidence drops below threshold
103  2. **Escalation zones** - Agent escalates if unsure (doesn't guess)
104  3. **Timeout limits** - Agent can't run forever
105  4. **Result validation** - Re-check after agent completes
106  
107  ```
108  Agent spawned → Runs with safeguards → Collapses if uncertain
109                                              ↓
110                                      Human reviews results
111  ```
112  
113  ---
114  
115  ## Axiom Alignment
116  
117  | Axiom | Alignment |
118  |-------|-----------|
119  | **A0 (Boundary)** | Thresholds define clear boundaries for action |
120  | **A1 (Integration)** | Automatic response integrates monitoring + action |
121  | **A3 (Navigation)** | Thresholds can be adjusted based on context |
122  | **A4 (Ergodicity)** | Prevents accumulation of issues that could compound |
123  
124  ---
125  
126  ## Instances
127  
128  ### Positive Instance: Speed Run Fixer Spawning
129  - **Context:** Speed run found 57 phoenix violations, 23 hygiene violations
130  - **Thresholds:** phoenix > 10, hygiene > 5
131  - **Action:** Automatically created 4 fixer agents
132  - **Result:** Fixers ready to execute without human having to count violations
133  - **Outcome:** ✓ Saved manual triage, same safeguards applied
134  
135  ### Positive Instance: Phoenix Hygiene Critical
136  - **Context:** LIVE-COMPRESSION.md staleness > 180 minutes
137  - **Threshold:** STALENESS_CRITICAL_MINUTES = 180
138  - **Action:** Exit code 2 (critical), blocks session start
139  - **Result:** Agent cannot proceed without addressing
140  - **Outcome:** ✓ Enforced hygiene without human remembering to check
141  
142  ---
143  
144  ## Related
145  
146  - [[validation-loop]] - After fixer runs, validate it worked
147  - [[infrastructure-over-suggestion]] - Prefer enforceable thresholds
148  - [[transparent-automation-principle]] - System acts, human overrides
149  
150  ---
151  
152  *proto-026 | Threshold-Triggered Automation | 2026-01-15*